If you've used an AI assistant to build something — a dashboard, a landing page, a quick internal tool — you've probably hit the same wall: the thing exists inside the AI session, and getting it in front of someone else is awkward. MCP is the reason that's changing.
Model Context Protocol, explained simply
MCP (Model Context Protocol) is an open standard that lets AI assistants connect to external tools and services. Think of it as USB for AI: a single, standardized plug that works across clients and tools, so nobody has to build custom integrations for every combination.
Before MCP, if you wanted an AI assistant to do something beyond text — check a database, send an email, deploy a page — someone had to build a bespoke integration between that specific AI and that specific service. MCP replaces that with a common protocol. Any AI client that speaks MCP can use any tool that speaks MCP.
The protocol was introduced by Anthropic and has since been adopted across the industry. It defines how an AI client discovers what tools a server offers, how it calls them, and how results come back — all over a standard transport (HTTP with server-sent events, or stdio for local tools).
Remote MCP servers vs. local ones
MCP servers come in two flavors, and the difference matters:
- Local MCP servers run on your machine. Your AI client starts them as a subprocess, usually via npx or uvx. They're great for things like file-system access or local database queries, but they only work when your machine is on, and they can't do anything that requires a public endpoint.
- Remote MCP servers run in the cloud. Your AI client connects to a URL — no subprocess, no local dependencies. They can handle things that need infrastructure: hosting, storage, email delivery, payments. They work from any device, and they survive when you close your laptop.
Sharing is a good example of something that requires a remote server. To give someone a link they can open in a browser, the content needs to be hosted somewhere public. A local MCP server can't do that — a remote one can.
How Bauta uses MCP to share what AI builds
Bauta is a remote MCP server. Connect it to any MCP-compatible AI client, and that client gains the ability to deploy HTML and React to live, shareable URLs. The AI doesn't just generate code — it publishes it to a real page that anyone can open in a browser.
The workflow is four steps, and only one of them is yours:
- Connect. Add Bauta's MCP endpoint (https://bauta.app/mcp) to your AI client. Authorize in the browser. This takes about a minute and you only do it once.
- Build. Ask your AI to build whatever you need — a dashboard, a report, a landing page, a tool.
- Share. Say "share this" (or "deploy this," or "publish this"). The AI calls Bauta's deploy_artifact tool and publishes the work to a live URL.
- Send the link. That's it. The recipient opens it in any browser. No installs, no accounts, no dev tools.
Which AI clients support MCP
MCP adoption has grown quickly. Here are the clients you can use with Bauta today:
| Client | Type | Setup |
|---|---|---|
| Claude (web & desktop) | Remote MCP connector | Add as connector |
| Claude Code | CLI agent | One terminal command |
| ChatGPT | Remote MCP connector | Add as connector |
| Codex | Cloud agent | MCP config |
| Cursor | Code editor | Settings UI or mcp.json |
| Windsurf | Code editor | Same MCP config as Cursor |
| Cowork | Team AI workspace | Plugin |
Same Bauta account, same URLs, same sharing controls — regardless of which client you use. Build in Claude, update from Cursor, adjust sharing from ChatGPT. The sharing layer doesn't care where the request came from.
Step by step: how MCP sharing works
Here's the concrete version. We'll use Claude as the example, but the flow is identical in any MCP-compatible client.
1. Connect Bauta
In your AI client, add Bauta as an MCP server. The URL is always the same:
When you connect for the first time, your client opens a browser window for authentication. Sign in or create an account — it's the same step. Authorization uses OAuth, so no API keys to copy around.
2. Ask the AI to build something
Ask for whatever you need. A client report. An interactive prototype. A comparison table. A landing page draft. The AI builds it as it always would.
3. Say "share this"
One sentence is enough. The AI calls Bauta's deploy tool, uploads the HTML (compiling JSX if needed), and returns a live URL:
4. Send the link
That URL works in any browser, on any device. The recipient doesn't need your AI client, doesn't need developer tools, doesn't need an account. If you update the content and republish, the same URL updates in place. Previous versions stay in version history.
Frequently asked questions
What is the Model Context Protocol?
Which AI clients support MCP?
Do I need to install anything?
Is my shared content public?
Try MCP sharing in two minutes
Connect Bauta to your AI client, build something, and say "share this." Free plan, no credit card, works with any MCP-compatible client.
Set up Bauta →